AWS Student Builder Group St. Pölten
Austria's only AWS Student Builder Group, run by students out of the University of Applied Sciences St. Pölten.
On the programme
Hybrid AI Architecture
2026 · St. Pölten
What actually happens when you wire local servers to the cloud: where the data has to stay, what each answer costs, and which of those two decides the architecture. Four ways to draw the same system, then three real projects on the layer their constraints forced.
OTTO: demo and presentation
2026 · St. Pölten
The multi-agent assistant, demonstrated and taken apart for the Cloud Club earlier in the year.
From the deck
Five slides, straight from the talks
Hybrid AI architecture: the same invoice job, four architectures. Slide from the deck.
What hybrid actually means: Bedrock in the Region, vLLM on OpenShift AI inside the building. Slide from the deck.
The video pipeline, the event assistant and the geospatial platform, side by side. Slide from the deck.
From MacBook to Bedrock: the agents and the models behind them. Slide from the deck.
Where the fine-tuned model runs, and what it costs. Slide from the deck.
The services doing the work
Read off the slides, not guessed. Each card says what that service did in this build, and links to the page explaining the capability and how to run it somewhere else.
Amazon Textract
The ready-made column in the invoice example: it returns the fields and you write the glue code, nothing else.
Amazon Bedrock
The ask-a-model column, and the regional half of the hybrid picture: one endpoint, one log, paid per token.
Amazon SageMaker AI
The train-your-own column: the same invoices, your model, behind your own endpoint.
Red Hat OpenShift AI
The run-the-platform-yourself column, and the ground inside the building in the hybrid split.
vLLM
Serves the model inside the building, for the data that may not leave it.
Amazon Bedrock AgentCore
Lets the agent reach on-premises data and tools while the inference stays in the Region. Only the agent crosses the line.
Amazon Bedrock Custom Model Import
System two: a 3B model tuned once on a rented GPU, then served managed. About 30 a month, almost all of it idle.
Amazon Rekognition
In the video pipeline: faces, shots and what is on screen.
Amazon Transcribe
The words, and when each one was said.
AWS Step Functions
Remembers where a long job got to, so a three-hour render survives a failure.
AWS Elemental MediaConvert
Renders the versions each platform wants.
Amazon API Gateway
In the geospatial platform: one front door, one audit trail, a key per tenant.
Amazon RDS
PostGIS on RDS holds the open spatial data in the cloud half. The survey and CAD files stay in an internal PostGIS.
AWS CloudTrail
Who called what, agents included, across two organisations operating two halves.
AWS IAM
A role per tenant, one identity each, nothing shared between customers.
InstructLab
The open-source pipeline behind the fine-tune: documents become Q&A pairs in YAML, synthetic data is generated from them, then the adapter is trained with LoRA. Model-agnostic, and it ran on a laptop.
CrewAI
Runs the crew: Q&A agent, agenda producer, header generator, speaker finder, GitHub planner and the UG Specialist.
AWS App Runner
Runs the agent container. The whole system came to about 30 euros a month.
Amazon ECR
Holds the container image App Runner pulls.
Amazon S3
Stores the trained model before Bedrock imports it.
Llama 3.1 3B
The open-weights base model that became the UG Specialist, adapted with LoRA rather than retrained.
Amazon Nova Pro
The agent model, and the teacher that generated the synthetic training data.
Claude Sonnet 3.7
Answers the open questions in the Q&A agent.
The longer write-ups
Where the builds behind these sessions are explained properly.
AI Architecture Layers: where a workload actually belongs
The longer write-up: the build, the services and what it costs to run.
Open-Source AI Wiki: Why I Built AI Solutions Wiki
The longer write-up: the build, the services and what it costs to run.
From AI Demo to Enterprise Solutions
The longer write-up: the build, the services and what it costs to run.
Enterprise AI Consultant vs. AI Consultant
The longer write-up: the build, the services and what it costs to run.
Cloud Day | FH Burgenland
The longer write-up: the build, the services and what it costs to run.
The same talk, other stages
Same session, a different room. Each page carries its own deck.
- AWS re:Invent
- AWS Community Day DACH
- AWS Community Day Slovakia
- AWS Community Day Czech Republic
- AWS Community Day Romania
- AWS Community Day Italy
- AWS Community Day Poland
- CTRL+ESC Malta
- Cloud Day FH Burgenland
Where all of this comes from
Every claim on this page traces back to the deck, the recap or the event’s own site.
- the group on Meetup, the event’s own site
- Slides: Hybrid AI Architecture, 2026
- Slides: OTTO: demo and presentation, 2026
- AI Architecture Layers: where a workload actually belongs
- Open-Source AI Wiki: Why I Built AI Solutions Wiki
- From AI Demo to Enterprise Solutions
- Enterprise AI Consultant vs. AI Consultant
- Cloud Day | FH Burgenland
Everything here is free to read, use and share. No email needed for the slides. If you reuse something, please credit the source.
This page mirrors the talk page on lindamohamed.com, where it was first published.